Hate speech is harmful content that directly attacks or promotes hatred against members of groups or individuals based on actual or perceived aspects of identity, such as racism, religion, or sexual orientation. This can affect social life on social media platforms as hateful content shared through social media can harm both individuals and communities. As the prevalence of hate speech increases online, the demand for automated detection as an NLP task is increasing. In this work, the proposed method is using transformer-based model to detect hate speech in social media, like twitter, Facebook, WhatsApp, Instagram, etc. The proposed model is independent of languages and has been tested on Italian, English, German, Bengali. The Gold standard datasets were collected from renowned researcher Zeerak Talat, Sara Tonelli, Melanie Siegel, and Rezaul Karim. The success rate of the proposed model for hate speech detection is higher than the existing baseline and state-of-the-art models with accuracy in Bengali dataset is 89%, in English: 91%, in German dataset 91% and in Italian dataset it is 77%. The proposed algorithm shows substantial improvement to the benchmark method.

A document by carlos. Click on the document to view its contents.

Intelligent nano-machines are a promising candidate technology for the next generation of health care. The realization of such units relies on novel, unconventional approaches, to allow for bio-compatibility and managing space constraints. In this work, we present three chemical processes, that can be used to realize a recently proposed molecular matrix multiplication unit on the lab-scale. The matrix multiplication is the fundamental operation for the realization of neural networks and, therefore, artificial intelligence. Hence, this work presents an important step towards practical realization of intelligent nano-machines for the next generation health care.

We always use timeline to describe time, but timeline unable to describe dynamic time. Because dynamic time is on changing, so timeline must be changing too, but it is impossible. Timeline includes layers, relationships existed between layers. If we changed a layer, the others need to be fixed too. Timeline can not make it up automatically, human being must take the work, therefore real-time-changing becomes impossible. It is no matter for film making, but internet needs instant responding, timeline can not cover it. Here we show a new structure called timetree, it is an auto-balanced hierarch structure. Its structure always be complete during changing without help from human being. It is a challenge for making dynamic interactive contents on internet, timetree is borned for it. We have tried timeline before, now it is the turn of timetree.

The advent of Knowledge-Based Planning (KBP) models has introduced a transformative approach to Intensity-Modulated Radiation Therapy (IMRT) treatment planning in breast cancer and lung cancer cases. This paper explores the application of KBP models in these specific cancer types, highlighting their potential to enhance treatment accuracy, efficiency, and patient outcomes. By leveraging historical treatment data and machine learning techniques, KBP-IMRT offers a data-driven framework for optimizing dose distributions, minimizing radiation exposure to healthy tissues, and improving overall treatment plan quality. Through a comprehensive review of the literature and clinical case studies, this paper underscores the advantages of KBP-IMRT, such as streamlined planning processes and improved plan consistency, while acknowledging the challenges associated with model development and implementation. As the field of radiotherapy continues to evolve, KBP models hold the promise of shaping the future of personalized and precise cancer treatment strategies.

As Generative AI becomes more prevalent, the vulnerability to security threats grows. This study conducts a thorough exploration of red teaming methods within the domain of Multimodal Large Language Models (MLLMs). Similar to adversarial attacks, red teaming involves tricking the model to generate unexpected outputs, revealing weaknesses that can be addressed through enhanced training for improved robustness. Through an extensive review of existing literature, this research categorizes and analyzes adversarial attacks, providing insights into their methodologies, targets and potential consequences. It further explores the evolving tactics employed to exploit vulnerabilities in various models, encompassing both traditional and deep learning architectures. The study also investigates the current state of defense mechanisms, examining countermeasures designed to thwart adversarial attacks. In addition to these aspects, the research conducts a meticulous analysis of red teaming methods with a specific focus on vulnerabilities related to images. By synthesizing insights from various studies and experiments, this survey aims to offer a comprehensive understanding of the multifaceted challenges posed by adversarial attacks in MLLMs. The outcomes of this research serve as a valuable resource for practitioners, researchers and policymakers seeking to fortify Generative AI systems against emerging security threats.

We discuss the crucial importance of explainability and understandability in artificial intelligence, in addition offering a small, insightful experiment, followed by a discussion of responses, challenges, and obstacles. We believe the pursuit of AI explainability and understandability is crucial, to be ignored at our peril.

We propose a branch-and-bound algorithm for robust rigid registration of two point clouds in the presence of a large number of outlier correspondences. For this purpose, we consider a maximum consensus formulation of the registration problem and reformulate it as a (large) maximal clique search in a correspondence graph, where a clique represents a complete rigid transformation. Specifically, we use a maximum clique algorithm to enumerate large maximal cliques and a fitness procedure that evaluates each clique by solving a least-squares optimization problem. The main advantages of our approach are i) it is possible to exploit the cutting-edge optimization techniques employed by current exact maximum clique algorithms, such as partial maximum satisfiablity-based bounds, branching by partitioning or the use of bitstrings, etc.; ii) the correspondence graphs are expected to be sparse in real problems (confirmed empirically in our tests), and, consequently, the maximum clique problem is expected to be easy; iii) it is possible to have a good control of suboptimality with a k-nearest neighbour analysis that determines the size of the correspondence graph as a function of k. The new algorithm is called CliReg and has been implemented in C++. To evaluate CliReg, we have carried out extensive tests on a dataset of 540 instances generated from scan-matching models in the public Standford 3D Scanning repository. The results show that CliReg clearly dominates the state-of-the-art (e.g., RANSAC, FGR, and TEASER++) in terms of robustness, with a running time comparable to TEASER++ and RANSAC. In addition, we have implemented a fast variant called CliRegMutual that performs similarly to the fastest heuristic FGR.

The standard solution to new technology is to center the ethics of robotics and artificial intelligence on ”concerns” of various kinds. Many of these fears end up being rather outdated; a few are essentially accurate but barely relevant (computer technological advances will annihilate businesses that make pictures on film, audio cassettes, or vinyl records); others are essentially accurate but extremely pertinent (automobiles will cause the deaths of children and drastically alter the landscape). Some of these fears are consistently incorrect when they indicate that technology will totally transform humans. This paper analyzes the problems and deflates the non-problems.

Natural language-based interfaces are frequently used in user contact with virtual assistants to overcome accessibility issues that some user groups may encounter. Although there are worries about how AI may affect jobs and possible biases in the use of virtual assistants, the technology has promise in a number of areas, such as everyday work support and mental health therapy. This paper evaluates the uses of AI in daily life in a comprehensive and informative manner.

The purpose of this paper is to examine how human beings and artificial intelligence might complement each other in management activities when it comes to higher productivity as a result of human-AI collaboration.

Deploying convolution-based algorithms into SRAM computing-in-memory (CIM) systems faces various challenges, such as operator incompatibility and intrinsic non-ideal error. This paper proposes a compilation framework to address this issue. Efficient weight mapping strategies are introduced to improve the utilization of SRAM-CIM macro. The intrinsic non-ideal errors of SRAM-CIM macro are also taken into consideration, and two efficient error correction schemes are proposed, which include calibration of computation voltage linear error (CCVLE) and the mitigation of analog-to-digital quantization error (MAQE). In addition, bit-width flexibility and signed-unsigned reconfigurability are also supported to facilitate the deployment of various convolution-based algorithms. ResNet18, finite impulse response (FIR) filtering, and Gaussian image filtering are deployed into a multi-macro SRAM-CIM system. These algorithms serve as deployment representatives of convolutional neural network (CNN), digital signal processing (DSP), and digital image processing (DIP), respectively. The results show that the introduced weight mapping strategies improve the macro utilization by 63.29% and 21.10% for two types of frequently used convolution layers compared to the traditional strategy. Moreover, the proposed error correction schemes achieve similar algorithm accuracy to the floating-point results, and the deployment result of ResNet18 achieves 66.3%~70.1% top-1 classification accuracy evaluated on the ImageNet dataset with different throughput tradeoffs.

This paper aims to broaden the horizons of symbolic artificial intelligence. The key to doing so is to establish a scientific basis for knowledge foundations, specifically for the concept of meanings and meaning computation. This led to the development of meaning computation artificial intelligence (MCAI) as a research and development framework providing a semantic modeling environment to deliver meaning computation intelligence solutions. MCAI is based on research in the areas of knowledge mathematics, semantic information structures, computational ontologies, and models of acquiring and applying knowledge. The paper discovered that relations appear to be fundamentally positioned at the core of knowledge. They are implicit relations (which reflect meaning) and semantic relations (which describe how implicit relations act between entities). As a result, knowledge mathematics was introduced based on algebras of implicit and semantic relations. On the basis of semantic information structures in the form of semantic triple chains and implicit relations algebra, the paper proposed a semantic knowledge model defined by an axiomatic computational model of relational domain ontology and an associated algebraic semantic reasoning method. The semantic knowledge model was successfully tested in the domains of time (including non-metric and metric interval relations) and space (spatial distance relations). Further, semantic relations algebra was applied to solving problems of knowledge graph intelligence. To accomplish this, a relational ontology-based model of knowledge graph was introduced. This led to the development of semantic reasoning models for resolving knowledge graph intelligence problems of completion and querying. Additionally, a novel intelligence task was introduced: knowledge graph compression, along with a semantic reasoning model for its resolution.

This paper focuses on the concept of convolution neural network for binary classification of food items. Two case studies were primarily focused : (i) identifying "rotten" and "fresh " food items using pre-trained models-VGG, ResNet50 and Xception; (ii) binary classification of "half-filled milk carton" and "completely filled milk carton" with different combination of selftrained neural network layers :dense layer, convolution layer and layer size combinations. In our first case study of binary classification, it was found that VGG had a validation accuracy of 97.5% and Xception produced an "overfitting" tendency. For our second case study, the combination with layer size =125, convolution layer = 3 and dense layer=1 had produced the highest validation accuracy of approximately 97 % and was also able to produce the most accurate prediction with different testing samples. This AI model can be implemented in smart refrigerators to let consumers know the status of their food item more accurately as compared to GoogleNet and NasNetLarge prediction.

Radio Access Networks (RANs) are responsible for the majority of the energy consumption of cellular networks, and their energy consumption has been growing at an unsustainable rate. Virtualized RANs (vRANs) are an alternative to monolithic RANs, which can adapt to ever-evolving traffic patterns in an agile way. Additionally, virtualization allows multiple radio sites to share the same computing infrastructure, which can lower energy consumption. However, the transition to vRANs is a slow process and there are many open questions about how to best manage vRANs to meet the requirements of cellular operators. In particular, it is unclear whether vRANs can reduce energy consumption compared to monolithic RANs and under which conditions. Energy models are crucial to address this question and identify opportunities for reducing energy consumption in vRANs. In this paper, we present an overview of the energy-modelling approaches that can be applied to RANs and we review models of monolithic and virtualized RANs. Additionally, we characterize the energy consumption of RANs at the network, node, component, and functional levels. Lastly, we examine the techniques that can be used to improve the performance of vRANs and highlight the challenges, unanswered questions, and opportunities in this field, concerning energy consumption.

In the context of India, a country with a rich tapestry of regional sign languages, effectively recognizing and interpreting Indian Sign Language (ISL) presents a formidable challenge for individuals with hearing and speaking impairments. This system introduces an innovative method for ISL recognition by leveraging the YOLOv5s (You Only Look Once version 5) object detection framework. Complementing the YOLOv5s, the project integrates Microsoft Azure’s cognitive app service, specifically the computer vision capabilities, and utilizes Mesa, a Python agent development framework. This comprehensive approach aims to enhance the expression and communication of individuals with hearing and speaking impairments in a predominantly spoken language-oriented world.

The k-anonymity problem introduced by Samarati and Sweeney in 1998, guarantees that it is impossible to distinguish user data from at least (k − 1) others in the same database. The methods used to achieve k-anonymity result in an information loss as the data in the database is modified, making it less accurate through a process of generalization or micro-aggregation of the stored data. Mauger et al. proposed a O(n²)-time sequential algorithm that gives good results while minimizing the information loss using their designed metrics. However, their solution is very time-consuming and therefore not suitable for large-scale databases. In this paper, we tackle this problem using parallelism. We propose three coarse-grained parallel algorithms to solve the k-anonymity problem. The first is the straightforward algorithm that runs in O(n ²/p) execution time with O(n) communication rounds, where n is the number of lines in the database and p is the number of processors. The second runs in O(n² /p²) execution time with O²(p) communication rounds. The third runs in O(n² /plog2(p)) execution time with O(np/ (log2(p))²) communications rounds. For the latter two algorithms, we introduce the concept of data reorganization to minimize the information loss when data are partitioned. Experimental results show that for a database of size n = 10^6 , p = 2^7 , and k = 10^2 , second, and third parallel algorithms are respectively 1127.59× and 41.13× faster than the sequential algorithm while achieving anonymity with 4.03% and 2.62% information loss.

In modern business, understanding customer behavior is of paramount importance. It allows for the delivery of personalized experiences, which can profoundly influence engagement and revenue. Behavioral segmentation is a potent approach known to enhance customer engagement by up to 73 percent and boost revenue by 15 percent. This study employs DBSCAN (Density-Based Spatial Clustering of Applications with Noise) to address the research problem of segmenting Amazon customers based on a wide range of features. This methodology enables a comprehensive analysis of customer behavior. The key findings of this study reveal customer segments at high risk of churn, a critical concern in the e-commerce industry. This insight is invaluable for businesses looking to tailor retention strategies, ultimately enhancing customer satisfaction and fostering sustainable growth within the fiercely competitive e-commerce landscape. The research showcases the practical application of DBSCAN for customer segmentation and churn risk analysis in the context of Amazon. Academics, managers, and decision-makers can leverage these findings to refine their strategies and better serve their customers, contributing to increased customer satisfaction and long-term business success.

The Artificial Intelligence for All initiative strives to ensure that artificial intelligence is accessible and beneficial to individuals from diverse backgrounds, regardless of their knowledge or expertise. Machine learning, a critical component of artificial intelligence, enables computers to learn from data and facilitates informed decision-making. Automated machine learning, or AutoML, automates the machine learning pipeline and thus makes it accessible to non-expert users. Although prior scientific research has reviewed various aspects of AutoML, its potential to democratize machine learning remains unexplored. In this survey, we aim to fill this gap by (i) reviewing systems and tools that help visualize the machine learning pipeline, (ii) evaluating the degree of user control in the process and (iii) identifying risks and limitations of current AutoML systems.

Recent studies in deep learning and neural networks have resulted in a significant breakthrough on several forefronts. Despite their success, the time and expertise required to build a neural network is immense. This led to the development of Automated Machine Learning (AutoML) which is a process to automate aspects of the machine learning pipeline. Neural Architecture Search (NAS) algorithm, a subset of AutoML focusing on architecture engineering, tends to be slow to train and is a computationally expensive process requiring a vast number of candidate networks which potentially limits their development. A solution to this problem would be to explore evaluation metrics that would allow us to get an estimation of a model’s performance while utilizing fewer computational resources. This paper is a compilation of all the recent developments in multiple sub-fields of AutoML, following which our discussion focuses on evaluation strategies that can provide intrinsic feedback about a proposed architecture without any traditional training while being computationally inexpensive. This will allow us to differentiate between good and bad performing architectures at initialization. We discuss the strategies employed by the current state-of-the-art, and their limitations, as we then propose multiple methods for running AutoML processes without training.

A binary up counter is presented, benefiting from the lower frequency characteristics of the higher-order bits to achieve high performance. The sequential circuit design is divided into two segments: a higher segment based on a slow counter design and a lower high-performance segment. The predictability and periodicity of directional counters is exploited to reduce the length of combinational paths to a minimum. This solution holds broad applicability for both ASIC and FPGA technologies. The design has been synthesized for two FPGA families: Intel Cyclone V and AMD UltraScale+. Comparative analysis against counter IP cores from Intel and AMD reveals that the proposed design consistently shows superior frequencies achieving up to a factor of 2.5× higher clock frequency having a resource consumption of approximately a factor 2× .

Analyzing videos presents a unique challenge due to their rich content compared to images. Furthermore, processing lengthy videos efficiently necessitates segmenting them into scenes. Focusing on individual scene analysis offers an efficient alternative to analyzing entire videos. The application of this approach extends to a variety of Video Intelligence tasks, from surveillance applications to comprehensive video analytics. By capitalizing on open-source foundation models and leveraging audio and text features, our framework offers a versatile solution to the intricate task of video analysis, catering to a multitude of real-world applications.  

Breast cancer is a pervasive global health concern among women. Leveraging multimodal data from enterprise patient databases-including Picture Archiving and Communication Systems (PACS) and Electronic Health Records (EHRs)-holds promise for improving risk assessment. This study introduces a multimodal deep-learning model leveraging mammogram datasets to evaluate breast cancer risk. Our approach integrates frozen large-scale pretrained vision-language models, showcasing superior performance and stability compared to traditional imagetabular models across two public breast cancer datasets. The model consistently outperforms conventional full finetuning methods by using frozen pretrained vision-language models alongside a lightweight trainable classifier. The observed improvements are significant. In the CBIS-DDSM dataset, the Area Under the Curve (AUC) increases from 0.867 to 0.902 during validation and from 0.803 to 0.830 for the official test set. Within the EMBED dataset, AUC improves from 0.780 to 0.805 during validation. In scenarios with limited data, using Breast Imaging-Reporting and Data System category three (BI-RADS 3) cases, AUC improves from 0.91 to 0.96 on the official CBIS-DDSM test set and from 0.79 to 0.83 on a challenging validation set. This study underscores the benefits of vision-language models in jointly training diverse image-clinical datasets from multiple healthcare institutions, effectively addressing challenges related to non-aligned tabular features. Combining training data enhances breast cancer prediction on the EMBED dataset, outperforming all other experiments. In summary, our research emphasizes the efficacy of frozen large-scale pretrained vision-language models in multimodal breast cancer prediction, offering superior performance and stability over conventional methods, reinforcing their potential for breast cancer risk assessment.

The presence of biases has been demonstrated in a wide range of machine learning applications, yet it is not yet widespread in the case of geospatial datasets. This manuscript illustrates the importance of auditing geospatial datasets for biases with a particular focus on disaster risk management applications, as lack of local data may direct humanitarian actors to utilize global building datasets to estimate damage and the distribution of aid efforts. It is important to ensure there are no biases against the representation of vulnerable populations and that they are not missed in the distribution of aid. This manuscript audits four global building datasets (Google Open Buildings, Microsoft Bing Maps Building Footprints, Overture Maps Foundation, and OpenStreetMap) for biases with regard to Relative Wealth Index, population density, urban/rural proportions, and building size in Tanzania and the Philippines. Dataset accuracies for these two countries are lower than expected. Google Open Buildings (with a confidence above 0.7) and OpenStreetMap demonstrated the best combinations of False Negative and False Discovery, though Google Open Buildings was more consistent across tiles. The equality of opportunity was lowest for the urban/rural proportions, whereas OpenStreetMap and Overture Maps Foundation displayed particularly low equality of opportunity for population density and RWI in Tanzania. These results demonstrate that there are biases in these geospatial datasets. The types of biases are not consistent across datasets and the two study areas which emphasizes the importance of auditing these datasets for biases for new applications and study areas.Note This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible.